in

Unsloth.ai Founder Q&A on Medium | #ArtificialIntelligence

Unsloth.ai Founder Q&A | Medium

In this “Unsupervised Learning” episode, Renee interviews Daniel from Unsloth, discussing their AI system that makes language model fine-tuning 30x faster. They explore Daniel’s AI journey, aiming for a personal ChatGPT for all, and discuss Unsloth’s future and common misconceptions about AI. Daniel and his brother founded Unsloth after participating in the LLM Efficiency Challenge hosted by Neurips, where they found training LLMs to be prohibitively slow and memory-hungry. They developed Unsloth to make fine-tuning 30x faster, use 70% less memory, and achieve +30% accuracy. The unexpected use of Unsloth was by people with no finetuning experience, who were able to use Unsloth via Google Colab notebooks. Unsloth is catering to people with no finetuning experience and has made the process simple. The biggest misconception about working with LLMs is thinking of them as a monolithic oracle, when in fact, the best way to approach LLMs is to stack them on top of each other. Unsloth is currently bootstrapped and people can support and contribute by donating through their Kofi account on Github. All their finetuning code is open source and freely available to run on a free Google Colab instance. Daniel keeps up with all the latest AI developments through Reddit, Twitter, Youtube, and Hacker News. Unsloth was intended to cater to hardcore machine learning engineers, but 50% of their audience turned out to be the average person who just wants to play with LLMs. Their goal is to bring the best of LLMs and the latest research to everyone. If Daniel had the chance, he would like to interview Ilya from OpenAI. He would like to talk to him about LLM’s ability to generalize, the double descent phenomena, and whether synthetic data might be the future. The source video and the full podcast with Daniel can be found on the Unsupervised Learning site.

Source link

Source link: https://reneeeshaw.medium.com/q-a-with-unsloth-founder-30x-faster-llm-fine-tuning-8191d957fcb5?source=rss——large_language_models-5

What do you think?

Leave a Reply

GIPHY App Key not set. Please check settings

My Waymo Autonomous Vehicle pulling up in SF! #waymo #selfdriving #aihype

#Waymo Autonomous Vehicle Arrives in San Francisco! #selfdriving #aihype

My First Waymo Autonomous Vehicle Ride with doggo! #waymo #selfdriving #aihype

#MyFirstWaymoRide with doggo! #waymo #selfdriving #aihype